[dmd/ctfeexpr] Don't copy ArrayLiteralExp.basis more than once

[ntrel]

Fixes #23367.

[thewilsonator]

Do we have a test case for this?

[dkorpel]

This would be an interesting PR for the benchmark CI that @abulgit is working on.

[ntrel]

@thewilsonator

Do we have a test case for this?

A self-contained example testing this is bug5147:
https://github.com/dlang/dmd/blob/master/compiler/test/compilable/interpret3.d#L805-L810

[ntrel]

A self-contained example testing this is bug5147:

Actually that one is strange, it sets basis but no elements are actually null. I think it's due to sarrayRepeat in initsem.d.

[ntrel]

With this PR, the following crashes:

alias f = {
    int[1][] pieces = [];
    pieces.length = 2;
    return pieces;
};
pragma(msg, f()); // segfault
static assert(f() == [[0], [0]]); // segfault

[ntrel]

I thought this might be an easy fix, but there's just too much code that ignores basis and indexes the elements field directly.

[WalterBright]

@ntrel your test case works now!

[thewilsonator]

Hmm, I might have messed up the rebase for this, the change is now trivial.